New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
speed up DA clusterizer allowing more vectorization #4435
Conversation
A new Pull Request was created by @VinInn (Vincenzo Innocente) for CMSSW_7_2_X. speed up DA clusterizer allowing more vectorization It involves the following packages: Alignment/OfflineValidation @nclopezo, @cmsbuild, @diguida, @rcastello, @StoyanStoynev, @slava77, @Degano can you please review it and eventually sign? Thanks. |
Working on it (for the above and for the record - see #4438). |
I'll have it finalized by tomorrow. At least I can say I have no issues with the code changes. There are a lot of (seemingly expected) differences seen in QCD and ttbar (with PU) samples that I need to sort out - all look a fluctuation ("random") type, no trend, acceptable. This is to finalize and also check timing. |
Timing. Tested on wf 202 (ttbar with PU; 200 events) and wf 38 (QCD, 1000 events). wf 25202 is consistent but has somewhat lesser precision (less events and less time per event for the interesting quantities). Overall - vertex times are affected positively (upto ~ 30+ %), some secondary vertex related quantities are affected negatively. wf 202 (note: should apply +10% for proper normalization): negatively affected: not affected within few % (for the record): In wf 38 (no PU) gains are much less pronounced (note: should apply +15% for proper normalization): combinedSecondaryVertexBJetTags 4.25616 ms/ev -> 7.02886 ms/ev particleFlowDisplacedVertex 16.0216 ms/ev -> 13.4798 ms/ev Event Size. It is affected including other objects further down the line but in a mild way "Branch Name | Average Uncompressed Size (Bytes/Event)": Comparison plots with summary will follow shortly. |
(red above is this PR) |
+1 %MSG-w TwoTrackMinimumDistance: PFTauPrimaryVertexProducer:hpsPFTauPrimaryVertexProducer 02-Jul-2014 21:03:13 CEST Run: 1 Event: 5088 It happened in one event only (though 77 times) out of 200 events. Tested 37e4476 on top of CMSSW_7_2_X_2014-06-27-0200 (as #4437, #4438, #4447). It is still mergeable in CMSSW_7_2_X_2014-07-04-0200 . |
This pull request is fully signed and it will be integrated in one of the next CMSSW_7_2_X IBs unless changes (tests are also fine). |
speed up DA clusterizer allowing more vectorization
35% speed up in the DA clusterizer itself at low pileup.
small regression observed in primary vertex reconstruction and other products whose producer algorithms depend on primary vertices (detached tracks, conversion)
https://twiki.cern.ch/twiki/bin/viewauth/CMS/HighGranularityCMSSWOptimization